AGPO (Adaptive Group Policy Optimization)

Modifies GRPO to handle two boundary cases: (1) when all $G$ responses in a group are uniformly correct or uniformly wrong ($\text{std}=0$), assigns a small non-zero signed advantage (+1 or -1) rather than 0 so the model still reinforces reliable completion or moves away from total failure, and (2) adds an adaptive length penalty to prevent redundant CoT inflation.